Wearable Devices in Cardiovascular Medicine

Photoplethysmography Detection of Irregular Rhythm

Time series analysis of PPG-derived pulse assessment can identify patterns in the pulse. Quantification of pulse rate variability or machine learning–based algorithms have been shown to successfully discriminate between sinus rhythm and atrial fibrillation using a variety of approaches.

Early approaches to identify atrial fibrillation were based on a conceptual framework similar to ambulatory ECG interpretation, which is to examine a 30-second interval of pulses. This was successfully performed with transillumination of the finger from a smartphone flashlight and detection by the adjacent camera. Eventually, the strategy was applied to watches. As the use cases expanded from fitness and wellness to diagnosis and disease management, these tools required greater regulatory oversight and clearance ( Fig. 12.2 ). An early attempt using third-party software on the Apple Watch had low specificity and positive predictive value. Subsequent approaches for irregular pulse identification were developed for high specificity, using a probabilistic approach of confirmatory pulse checks over hours or days. An algorithm designed for the Apple Series 1, 2, and 3 watches (Apple Inc, Cupertino, CA) intermittently and passively measures pulse over 1 minute to generate a beat-to-beat pulse tachogram ( Fig. 12.3 ). If this tachogram meets irregularity criteria, then the algorithm temporarily increases the sampling frequency. If five out of six consecutive tachograms met irregularity criteria, then the algorithm notifies the user of an irregular rhythm. Therefore, unlike the classical ECG definition of AF, which requires a consecutive duration of only 30 seconds, this PPG-based algorithm is probabilistic, requiring multiple episodes to meet criteria (see Fig. 12.3 ), and is therefore considered less sensitive, especially for very short AF episodes, but much more specific.

The Apple algorithm was tested at scale in a single-arm, unblinded, investigational device exemption study. Inclusion criteria included age ≥22, possession of compatible Apple watches and phones, no prior history of AF, and U.S. residency. Over an 8-month period between 2017 and 2018, the study enrolled 419,297 U.S. participants. Overall 0.52% of participants received an irregular rhythm notification. Among 450 participants with notifications who received ambulatory ECG patch monitoring, AF was a detected on that patch in 34% (97.5% CI 29% to 39%). The positive predictive value for an irregular rhythm notification was 0.84 (95% CI 0.76 to 0.92). Because only notified participants received “gold standard” ECG monitoring, the study was unable to assess sensitivity or specificity.

Studies similar in design have been performed to test similar PPG-based algorithms on other smartwatch platforms. The Huawei Heart Study enrolled 246,541 participants in China to evaluate a fitness band and smartwatch and had directionally similar results, although a lower proportion receiving notifications, possibly due to a younger population or greater algorithm specificity. In May 2020, Fitbit launched its own study to evaluate an algorithm on their device platform with target enrollment of 100,000 participants ( https://clinicaltrials.gov/ct2/show/NCT04380415 , accessed September 5, 2020).

Although these studies indicate the promise for undiagnosed AF detection in an at-risk population, the FDA views this class of algorithms as prediagnostic tools, rather than serving as more definitive diagnostic tests. On the Apple platform, consumers must opt in to enable these features on their watches. In doing so, they receive onboarding that includes education. Because the sensitivity of these tests is not known and because they are enabled by the user rather than by a clinician as a public health intervention, these tools do not meet the classical Wilson-Jungner criteria for screening tests. Presently, there are no professional society or U.S. Preventative Services Task Force recommendations for their use for AF surveillance, screening, or diagnosis.

Electrocardiogram

More recent smartwatch models (Apple Watch Series 4 or higher, Samsung Galaxy Watch 3) have FDA-cleared single-lead ECG capability (see Fig. 12.1 ). The user actively records a 30-second lead I (right arm [−] to left arm [+]) ECG on the watch by pressing the crown with a finger of the hand opposite the hand with the watch body electrode. However, the first major smartphone-connected ECG was released in 2013 (AliveCor, Mountain View, CA). The Kardia device has two electrodes (one for each hand) and communicates wirelessly to a smartphone. A new six-lead consumer version (Kardia 6L) is now available that uses the right leg for additional limb and derived ECG leads.

Consumer-based ECG devices entail substantial limitations. Compared with medical 12-lead systems and patch-based ECG monitors, smartphone-connected and smartwatch ECG devices tend to have significantly more artifact. To counter this, aggressive filtering and baseline drift correction may be applied, which may obscure important ECG features. For example, a smartwatch algorithm incorrectly labeled a tracing of atrial tachycardia as atrial fibrillation ( Fig. 12.4 ). Careful review shows that atrial tachycardia P waves have been attenuated due to filtering, which was clearly present on the medical 12-lead ECG. However, ST changes during acute coronary syndromes as measured by the Apple Watch in all 12 lead positions have shown good agreement with medical grade systems.

Another major caveat is that consumer-based ECG systems also do not provide comprehensive prediagnostic information across a variety of rhythms compared with medical grade systems. Numerous examples of incorrect diagnosis, including of sustained ventricular arrhythmias, have been documented. Moreover, a ventricular rate that is out of range (<50 or >100) prevents automated ECG interpretation on the Apple Watch.